Time series and forecasting · insight-0211
Treat intermittent demand forecasts as rates and use bias-corrected Croston-type methods
- Evidence type
- documented fact
- Difficulty
- medium
- Roles
- Data scientist
- Industries
- manufacturing, aviation, retail
Summary: situation, action, reason, boundary
Situation
A spare-parts series is zero most weeks with occasional orders of 3 to 20 units, and exponential smoothing chases each spike.
Action
Benchmark Croston-type methods that smooth demand size and inter-demand interval separately, preferably a bias-corrected variant such as SBA, against a simple moving average, and use the result as a per-period demand rate.
Reason
Croston separated size and interval for intermittent series; Syntetos and Boylan showed the original estimator is biased and evaluated a correction.
Boundary
Spare parts, slow-moving retail SKUs, aircraft rotables, low-volume routes.
Symptom or risk
Standard smoothing overreacts after each order and decays slowly, producing a biased, jumpy forecast that drives poor reorder points.
Exceptions
If demand is lumpy and driven by known orders or projects, a judgmental or order-book forecast may beat any time-series method.
Trade-offs
Rate forecasts do not say when demand arrives; inventory decisions need the full lead-time demand distribution.
Pitfalls
Evaluating with MAPE, which is undefined on the zero weeks.
Lessons and concepts
Claims and their sources
Croston's method separately smooths non-zero demand sizes and the intervals between demands.
Croston's estimator is biased, and the bias-corrected Syntetos-Boylan Approximation was proposed and evaluated.